Recognition of human fingers shape for control of humanoid robotic hand

摘要

The goal of this study is to construct a human-robot communication system. In the present study, therefore, we propose a technique to recognize motions of human hand and fingers from a single COD image sequence, so as to extract motion parameters and control a humanoid robotic arm with hands. It is possible from the image to recognize the shape and posture of human hand and fingers if it is of a rigid body with simple shape. But a joint object with combination of rigid bodies may generally need both geometric model constructed beforehand and matching the images with the model. In this paper, we decrease computational complexity by adopting the range limitation of finger angles and the minimum accuracy for the control. So we estimate the fingers' shape using self-organizing feature map. we decided the control parameter of humanoid robotic hand from the estimated result of the fingers' shape. So, we carried out control experiments of humanoid robotic hand with these control parameters.